Detection of Sesame Disease Using a Stepwise Deep Learning Approach

• Main Authors: Abeje, Bekalu Tadele, Salau, Ayodeji Olalekan, Tadesse, Esayas Gebremariam, Ayalew, Aleka Melese
• Format: Conference Proceeding
• Language: English
• Subjects: Cameras; Convolutional Neural Network; Crops; Deep learning; Disease detection; Feature extraction; Image processing; Image segmentation; Sesame disease; Technological innovation; Training
• ISSN: 2770-7466
• DOI: 10.1109/3ICT56508.2022.9990780

Sesame, along with coffee, is Ethiopia's most exported product and the country's main source of foreign exchange. It is mostly cultivated in the northern parts of Ethiopia. The products derived from sesame vary from year to year due to different factors. Among the factors are weather condition and disease. Sesame crops are susceptible to a variety of diseases, and this is one of the main reasons why farmer output is declining. The several disease kinds that result from a large sesame farm are difficult to distinguish with the necked eye. Therefore, in this paper, we propose a stepwise deep convolutional neural network approach to easily identify sesame disease. In the proposed approach, image-processing steps mainly image acquisition, preprocessing, segmentation, data augmentation, feature extraction, and classification were considered. Images were collected from the northern part of Ethiopia, mainly the Amhara region using Samsung A32 and iPhone 6s phone cameras with a 450x680-pixel resolution. 540 infected plant images were collected of Bacteria Blight, Phyllody, and Healthy plants to enable the convolutional neural network to extract important features from the segmented images. Finally, SoftMax fully connected layers was employed to classify the images into their respective classes of sesame disease. The proposed model achieves 99% training accuracy and 98% testing accuracy.